svm_fifo rework to avoid contention on cursize

Problems Addressed:
- Contention of cursize by producer and consumer.
- Reduce the no of modulo operations.

Changes:
- Synchronization between producer and consumer changed from cursize
  to head and tail indexes
  Implications: reduces the usable size of fifo by 1.
- Using weaker memory ordering C++11 atomics to access head and tail
  based on producer and consumer role.
- Head and tail indexes are unsigned 32 bit integers. Additions and
  subtraction on them are implicit 32 bit Modulo operation.
- Adding weaker memory ordering variants of max_enq, max_deq, is_empty
  and is_full Using them appropriately in all places.

Perfomance improvement (iperf3 via Hoststack):

iperf3 Server: Marvell ThunderX2(AArch64) - iperf3 Client: Skylake(x86)
   ~6%(256 rxd/txd) - ~11%(2048 rxd/txd)

Change-Id: I1d484e000e437430fdd5a819657d1c6b62443018
Signed-off-by: Sirshak Das <sirshak.das@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>

2 files changed, 14 insertions, 13 deletions

diff --git a/src/vcl/vcl_private.c b/src/vcl/vcl_private.c
index a572fd4d540..3e82becfc93 100644
--- a/src/vcl/vcl_private.c
+++ b/src/vcl/vcl_private.c
@@ -346,9 +346,9 @@ vcl_session_read_ready (vcl_session_t * session)
     return clib_fifo_elts (session->accept_evts_fifo);
 
   if (vcl_session_is_ct (session))
-    return svm_fifo_max_dequeue (session->ct_rx_fifo);
+    return svm_fifo_max_dequeue_cons (session->ct_rx_fifo);
 
-  return svm_fifo_max_dequeue (session->rx_fifo);
+  return svm_fifo_max_dequeue_cons (session->rx_fifo);
 }
 
 int
@@ -365,7 +365,7 @@ vcl_session_write_ready (vcl_session_t * session)
   if (PREDICT_FALSE (session->session_state & STATE_LISTEN))
     {
       if (session->tx_fifo)
-	return svm_fifo_max_enqueue (session->tx_fifo);
+	return svm_fifo_max_enqueue_prod (session->tx_fifo);
       else
 	return VPPCOM_EBADFD;
     }
@@ -383,9 +383,9 @@ vcl_session_write_ready (vcl_session_t * session)
     }
 
   if (vcl_session_is_ct (session))
-    return svm_fifo_max_enqueue (session->ct_tx_fifo);
+    return svm_fifo_max_enqueue_prod (session->ct_tx_fifo);
 
-  return svm_fifo_max_enqueue (session->tx_fifo);
+  return svm_fifo_max_enqueue_prod (session->tx_fifo);
 }
 
 /*
diff --git a/src/vcl/vppcom.c b/src/vcl/vppcom.c
index 037df995510..b5953872f81 100644
--- a/src/vcl/vppcom.c
+++ b/src/vcl/vppcom.c
@@ -1496,14 +1496,14 @@ vppcom_session_read_internal (uint32_t session_handle, void *buf, int n,
   rx_fifo = is_ct ? s->ct_rx_fifo : s->rx_fifo;
   s->has_rx_evt = 0;
 
-  if (svm_fifo_is_empty (rx_fifo))
+  if (svm_fifo_is_empty_cons (rx_fifo))
     {
       if (is_nonblocking)
 	{
 	  svm_fifo_unset_event (s->rx_fifo);
 	  return VPPCOM_EWOULDBLOCK;
 	}
-      while (svm_fifo_is_empty (rx_fifo))
+      while (svm_fifo_is_empty_cons (rx_fifo))
 	{
 	  if (vcl_session_is_closing (s))
 	    return vcl_session_closing_error (s);
@@ -1527,7 +1527,7 @@ vppcom_session_read_internal (uint32_t session_handle, void *buf, int n,
   else
     n_read = app_recv_stream_raw (rx_fifo, buf, n, 0, peek);
 
-  if (svm_fifo_is_empty (rx_fifo))
+  if (svm_fifo_is_empty_cons (rx_fifo))
     svm_fifo_unset_event (s->rx_fifo);
 
   VDBG (2, "session %u[0x%llx]: read %d bytes from (%p)", s->session_index,
@@ -1577,14 +1577,14 @@ vppcom_session_read_segments (uint32_t session_handle,
   if (is_ct)
     svm_fifo_unset_event (s->rx_fifo);
 
-  if (svm_fifo_is_empty (rx_fifo))
+  if (svm_fifo_is_empty_cons (rx_fifo))
     {
       if (is_nonblocking)
 	{
 	  svm_fifo_unset_event (rx_fifo);
 	  return VPPCOM_EWOULDBLOCK;
 	}
-      while (svm_fifo_is_empty (rx_fifo))
+      while (svm_fifo_is_empty_cons (rx_fifo))
 	{
 	  if (vcl_session_is_closing (s))
 	    return vcl_session_closing_error (s);
@@ -1681,14 +1681,15 @@ vppcom_session_write_inline (uint32_t session_handle, void *buf, size_t n,
   is_ct = vcl_session_is_ct (s);
   tx_fifo = is_ct ? s->ct_tx_fifo : s->tx_fifo;
   is_nonblocking = VCL_SESS_ATTR_TEST (s->attr, VCL_SESS_ATTR_NONBLOCK);
+
   mq = wrk->app_event_queue;
-  if (svm_fifo_is_full (tx_fifo))
+  if (svm_fifo_is_full_prod (tx_fifo))
     {
       if (is_nonblocking)
 	{
 	  return VPPCOM_EWOULDBLOCK;
 	}
-      while (svm_fifo_is_full (tx_fifo))
+      while (svm_fifo_is_full_prod (tx_fifo))
 	{
 	  svm_fifo_add_want_tx_ntf (tx_fifo, SVM_FIFO_WANT_TX_NOTIF);
 	  if (vcl_session_is_closing (s))
@@ -2003,7 +2004,7 @@ vppcom_select (int n_bits, vcl_si_set * read_map, vcl_si_set * write_map,
         continue;
       }
 
-    rv = svm_fifo_is_full (session->tx_fifo);
+    rv = svm_fifo_is_full_prod (session->tx_fifo);
     if (!rv)
       {
         clib_bitmap_set_no_check ((uword*)write_map, sid, 1);